Study population
The present work utilized data of four independent cohorts that included patients with cerebrovascular disease on whom DCS monitoring was performed using similar HOB protocols. One cohort of patients had asymptomatic severe (at least 70%) ICA stenosis or occlusion and healthy controls. The other three cohorts were comprised of patients with AIS.
Two of the AIS cohorts were studied at the Hospital of the University of Pennsylvania, USA: (1) from 2005 to 2007 [17] (PENN05–07) and (2) from 2009 to 2011 [16] (PENN09–11). Two other studies were conducted at the Stroke Unit of Hospital de la Santa Creu i Sant Pau of Barcelona: (3) a study on AIS patients from 2015 to 2017 [20] (BCN15–17), and (4) a study on ICA stenosis patients and healthy volunteers (BCN-study), both with the same HOB manipulation protocol. All protocols were approved by local internal review boards, and the participants or their legal proxies provided written consent to participate. All methods were performed in accordance with the relevant guidelines and regulations.
PENN05–07, PENN09–11 and BCN15–17 studies included patients admitted to the stroke service with AIS affecting the anterior circulation. The general exclusion criteria were intracranial hemorrhage on initial neuroimaging, and inability to lie supine for 15 min.
The BCN-study included asymptomatic patients (defined as no stroke in the territory of the stenotic artery in the preceding 6 months) with severe unilateral or bilateral extracranial ICA stenosis (at least 70%). The latter were referred to the neurosonology lab for cerebrovascular reserve testing. The exclusion criteria were bilateral inadequate temporal acoustic windows for sufficient TCD examination or the evidence of an additional intracranial stenosis in the anterior circulation. Neurologically healthy volunteers were also included in the study as controls.
The specific inclusion and exclusion criteria for each study can be found on the corresponding publications [16, 17, 20] and in Supplementary Material.
Head-of-bed manipulation protocol
Different HOB position alteration protocols are illustrated in Fig. 1. All protocols involved orthostatic challenges at different HOB positions including repeated supine positions (highlighted with circles in Fig. 1). For all studies, optical data was acquired for 5 min at each position. The transition between HOB positions was noted as event markers in the data.
AIS patients were placed flat with the HOB between 0° and 15° according to the local clinical practice guidelines during the first 24 h after the presumed stroke onset. Afterwards, mobilization was guided by the judgment of the attending clinician.
For the PENN05–07 and PENN09–11 studies [16, 17], the study protocol was planned for three separate days and the first measurement was performed as soon as the patient was available. For the BCN15–17 study [20], the study protocol was initiated within the first 48 h after symptom onset. The protocol was repeated up-to four times at intervals of 48 h during the first week of admission as long as the patient was stable. For the BCN-study (on patients with ICA stenosis and controls), the measurements were performed once.
Optical methods and instrumentation
The specific optical methods and instrumentation for the measurement of CBF and MAP for each study can be found in previous publications [16, 17, 20] and in Supplementary Material. Here we briefly outline the salient, common features.
For all studies, two non-invasive, optical probes were placed on the forehead bilaterally and as laterally as possible to avoid the frontal sinuses. This selection was guided by practical reasons (to avoid hair) and also since previous studies by other methods [33] and, by DCS [32], have shown that the frontal cortical area is a valid area of measurement when studying the global cerebral vasoreactivity (CVR) at the level of individual cerebral hemispheres. The probes consisted of detector fibers and a source fiber set at 2.5 cm from the detector fibers. A 2.5 cm separation provides information about the cortical cerebral hemodynamics, as previously validated [32, 34, 35].
Diffuse correlation spectroscopy has been extensively validated for measuring relative microvascular cortical CBF against other modalities [32, 35]. Recently, Giovannella et al. [36] has paved the way to calibrate for accurate absolute CBF measurements by DCS on neonates which would be applicable for adult brain measurements with appropriate means to account for the partial volume effects. Finally, we stress that, unlike near-infrared spectroscopy (NIRS), DCS is a direct measure of CBF. NIRS, on the other hand, provides surrogate measures of CBF by making assumptions about oxygen extraction and blood volume and their relationship to CBF.
The DCS system [31] employed a long coherence length laser (785 nm) single photon avalanche photo-diode detectors and a hardware auto-correlator. DCS evaluates the statistics of the diffuse laser speckles by using the auto-correlation function of the detected light intensity fluctuations. The blood flow index (BFI) of the local microvasculature is then calculated by fitting the appropriate solution of the correlation diffusion equation to the intensity autocorrelation function as previously described [30].
The changes in CBF at the second supine position were calculated by using the mean BFI at the first supine position as the baseline. The results are reported as ΔrCBF = \( \overline{\left(\left(\frac{BFI_{supine2}(t)}{\overline{BFI_{supine1}}}-1\right)x100\right)} \), where \( \overline{BFI_{supine1}} \) is the average of the cerebral BFI during the first supine position and BFIsupine2(t) is the continuous BFI data during the measurement at the second supine position. Up-to 1 min of the continuous BFI data was discarded for each HOB position in order to avoid bed movement artifacts.
MAP was measured continuously or at half-time of each HOB position. For the continuous MAP measurements, the first and last minutes from the analysis were discarded and the rest were used to calculate a mean of each head-of-bed (HOB) position. These data were obtained continuously by a non-invasive blood pressure monitor Finapres (Finapres Medical Systems, Arnhem, the Netherlands) device in a sub-set of patients. The mean of the MAP changes is reported as ΔMAP = \( \overline{\left({MAP}_{supine2}(t)-\overline{\left({MAP}_{supine1}\right)}\right)} \). \( \overline{\left({MAP}_{supine1}\right)} \) is the average of the MAP during the first supine position and MAPsupine2(t) is the continuous MAP data during the measurement on the second supine position.
When the measurements were performed at half-time of each HOB position, a manual sphygmomanometer (Omron BP785 IntelliSense Automatic Blood Pressure Monitor, Omron, Osaka, Japan) was used to measure the MAP at 2.5 min from each HOB position change. The changes are reported in the same manner as the continuous measurements.
Clinical and imaging evaluation
The baseline examinations included the collection of demographics and vascular risk factors and a physical examination obtained by certified neurologists or senior residents under supervision who were blinded to the optical information. Diabetes mellitus, arterial hypertension and dyslipidemia were obtained for all stroke studies.
The etiologic stroke subtype was classified according to the modified Trial of Org 10,172 in Acute Stroke Treatment (TOAST) [37] criteria in two stroke studies. The extent of early ischemic changes was evaluated by the Alberta Stroke Program Early Computed Tomography Score (ASPECTS) [38] in two stroke studies. The specific clinical and imaging evaluations for each study can be found in previous publications [16, 17, 20] and in Supplementary Material.
Statistical analysis
Quantitative clinical variables are described as a median and an interquartile range (IQR) and categorical variables as number of cases and percentages of the total (cases (percentages)). Demographic characteristics were compared across groups using either the Kruskal-Wallis test (for quantitative variables) or Fisher’s exact test (for categorical variables). If the global test was statistically significant, multiple pairwise comparisons were made using the Wilcoxon rank-sum or Fisher’s exact test to assess differences with adjustment using Holm-Bonferroni correction.
For the patient groups, each cerebral hemisphere was tagged as being “ipsilesional” or “contralesional” based on the presence of the pathology on that hemisphere. In the case of AIS patients, “ipsilesional” refers to the cerebral hemisphere where the hemisphere with cerebral ischemia was observed. In the case of ICA stenosis subjects, the categorization was by degree of the asymptomatic ICA stenosis as severe (≥70% or occlusion, “ipsilesional”) and non-severe (stenosis< 70% or absent, “contralesional”). Only patients with unilateral ICA were considered when evaluating the associations between CBF and MAP.
There are no known pathological lesions present for the healthy volunteers, we have thus randomly assigned each hemisphere measured in the controls as “side 1” or “side 2”. This was done to avoid any systematic bias by using the left/right indication.
The AIS group involved repeated measurements. In this case, linear mixed-effect models (if lack of independence in the response variable) were used for checking if the mean response of ΔrCBF or ΔMAP differed from zero, where patient identifier, the study name and the hemisphere were, if needed, the random factors for these analyses. Otherwise, simple linear models were used when the response variable was independent. Linear mixed-effect models (if there was a lack of independence in the response variable) were also used to study the association between ΔrCBF and ΔMAP, where patient identifier was the random factor. If needed in the specific model, the hemisphere and/or the study name were the covariables. Again, simple linear models were used when the response variable was independent.
For the linear models we report estimates of the mean effect along with 95% confidence intervals (95% CI). The p-values are reported for the hypothesis test of whether the effect is zero. Significance of specific terms was assessed using a likelihood ratio test for the full versus reduced models. A type I error of 0.05 was used to accept significance without adjustment for multiple comparisons. The “nlme” software package was used for the linear mixed-effect models implemented in R [39]. All statistical analyses were performed with R [39].